New York AI Ethics Rules for Lawyers

The official instruments

Three layers, one workflow

No jurisdiction has built more AI architecture than New York, and none rewards a layered reading more. The layers differ in authority: a bar association report (persuasive policy), city bar formal opinions (the operative ethics analysis for practitioners statewide, since New York has no state-level AI opinion), and a court rule (binding). Conflating them is the most common New York error; this page keeps them separate.

Layer one: the NYSBA Task Force report (April 6, 2024). Adopted by the House of Delegates, the report examines AI’s legal, social, and ethical impact, keys guidelines to the Rules of Professional Conduct, recommends education over new legislation, and proposes a standing committee to keep guidance current. It also advises disclosure of AI use to clients, a position stated more flexibly in the later opinions. Treat it as the policy frame and a drafting resource, not as a compliance standard; it is labeled a report on the tracker for exactly that reason.

Layer two: the NYC Bar opinions. Formal Opinion 2024-5, dated August 7, 2024 (a date worth being precise about, since April 2024 belongs to the NYSBA report and the misdating circulates widely), is the operative analysis: generative AI needs guardrails, not new rules. Its holdings run the full board: protect confidences before they enter tools, with informed consent for open systems; verify outputs and citations, citing Mata v. Avianca and United States v. Cohen; supervise firm-wide use; bill actual time only. Its most practice-shaping move is the exemption: routine embedded AI (autocomplete, the features inside established research platforms) requires no disclosure, which keeps the duty pointed at genuine perimeter-crossing uses. Formal Opinion 2025-6 (December 22, 2025) then extends the framework to the tool category firms adopted fastest: AI meeting recorders and transcribers, requiring informed client consent before recording client conversations, vendor vetting, and transcript accuracy checks.

Layer three: Part 161, the binding one. Adopted March 25, 2026, effective June 1, 2026, Part 161 covers every UCS court in civil and criminal cases and takes the accountability-over-disclosure position: AI use in preparing court papers is permitted, and no system-wide disclosure is required. Its Appendix A is the part to operationalize: an opt-in model rule for individual courts under which the signature on a paper certifies careful review and independent confirmation that it contains no fabricated cases, statutes, or other material, enforceable through Part 130 sanctions and RPC 3.3. Evidence is excluded from “papers.” The rule converts the verification workflow from ethics best practice into the literal meaning of your signature wherever Appendix A is adopted, and the prudent firm treats it as adopted everywhere.

Practicing under the full stack

The composite New York workflow: vet tools and record terms (2024-5’s open-system line decides where consent is needed); put standing AI language in engagement letters and obtain consent before client conversations are recorded (2025-6) or confidences enter open systems (2024-5); verify every citation in a traditional database before filing, then check whether the specific court adopted Appendix A and whether the judge keeps an individual standing order, both via the court’s own pages with the Ropes & Gray tracker as cross-check; bill actual time. Multi-layer also means multi-speed: the court rule is new and courts’ Appendix A adoptions will accumulate, which is precisely the kind of movement the changelog and alerts track. For the duties in firm-policy form, the template maps section-by-section onto everything above.